Releases: FluidNumerics/SELF
Releases · FluidNumerics/SELF
Version 0.0.1
Overview
This is the first official version release of SELF. It provides the code structure necessary for implementing the Discontinuous Galerkin Spectral Element Method for generic hyperbolic and parabolic conservation laws in 1-D, 2-D, and 3-D on CPU only, Nvidia GPU accelerated, and AMD GPU accelerated hardware.
Feature Set
- Verified support for AMD MI210 and x86 architectures with a range of GNU compilers.
- Classes for working with mathematical scalars, vectors, and tensors 1-D, 2-D, and 3-D with isoparametric geometry on unstructured grids
- Abstract Model Class that provides Euler and Runge-Kutta series explicit time integrators for models defined through type extensions
- Base classes for model implementation in 1-D, 2-D, and 3-D; these base classes define a generic forward stepping pipeline that can be used for conservation law modeling. Because we use OO Fortran, end users are free to override these procedures to customize the forward-stepping pipeline to their heart's content.
- Example linear hyperbolic models, including shallow water (2-D) and linear euler equations in 2-D and 3-D. Example nonlinear Burgers equation solver is provided in 1-D.
Alpha Release of Version 0.0.1
Version 0.0.1
This release has been verified on a limited number of platforms to demonstrate algorithm correctness for core spectral element algorithms implemented in the SELF_Lagrange
, SELF_Data
, SELF_Mesh
, SELF_Geometry
, and SELF_MappedData
modules.
Verified Platforms
- GNU 9.2.0, ROCm 4.1.0 - AMD MI50 (gfx906), AMD MI100 (gfx908)
- GNU 10.2.0, ROCm 4.0.1, CUDA 11.1 - Nvidia Tesla v100 (sm_70)